/

En

PRCV2020 Challenge(1): Large-scale Pedestrian Retrieval


1. The purpose and significance of the competition

Pedestrian retrieval can help users finding interesting pedestrian targets appearing in specific videos and images according to query conditions (such as attribute values ​​or pictures). It is one of the important terminal applications of intelligent visual monitoring systems and has strong practical value in the field of public safety. At present, researches related to pedestrian retrieval, such as pedestrian re-recognition and pedestrian attribute recognition, are developing rapidly. There are many researchers at home and abroad, and there are many articles and methods. However, most of the research is mainly focused on the algorithm level, trying to solve the many technical difficulties (such as low resolution, occlusion, and posture changes) faced by pedestrian attribute recognition or pedestrian re-recognition models to provide algorithm robustness and recognition Precision. At present, there is still a lack of performance evaluation standards and test methods for "end-to-end" application systems from original video to pedestrian retrieval. In view of the above, this competition will provide a benchmark for large-scale pedestrian retrieval performance evaluation for real monitoring. On the one hand, it is used to investigate the recognition performance of pedestrian attribute recognition and pedestrian re-recognition algorithms related to pedestrian retrieval under different target scales, posture changes, and occlusion conditions; at the same time, it will also provide a large-scale video analysis platform-ISEE system, It is used to examine the system performance of the pedestrian attribute recognition and pedestrian re-identification algorithm model and the pedestrian detection algorithm model. This will more directly examine the performance of the pedestrian retrieval system in the form of "end-to-end", which will help to find that the integration of the algorithm affects the overall system. Performance bottlenecks, explore the impact of large-scale interference samples in the query set on retrieval performance, and form evaluation criteria and test methods for pedestrian retrieval systems.

2. Organization

Organizer: CASIA, Shandong University of Science and Technology
Contact: Teacher Zhang Zhang, zzhang@nlpr.ia.ac.cn; Teacher Caifeng Shan, caifeng.shan@gmail.com
Sponsor: Genjian Intelligent Technology Co., Ltd.

3. Requirements

Research institutions at home and abroad, including various universities, research institutes, and company research institutes, can participate.

4. Way of registration

Only the team registration is accepted by mail, each team has no more than 5 people, and each contestant can only participate in 1 team. The participating teams signed the competition agreement and stamped by the unit, and the scanned version should be sent to the competition contact mailbox da.li@cripac.ia.ac.cn.
  • Message header format: “PRCV2020 Large-scale Pedestrian Retrieval competition + team name”.
  • The content of the email includes: basic information of the team; information of the person in charge; number of participants; contact information (mobile phone number and email); content of the competition; the scanned versiion of competition agreement which is stamped by the unit.
  • Registration deadline: July 15, 2020.
After the organizer received the email and confirmed with the applicant, the registration was successful.

5. Competition schedule(Beijing time)

  • May 01 - July 15: Registration and the confirmation of qualifications(the deadline is 23:59, July 15).
  • May 31 - July 15: Open data set (available for download in the form of a network disk, the download link is sent to the reserved mailbox of the participating team by email, please complete the data download as soon as possible during the data opening time, and please contact the organizer in time if you encounter problems).
  • June 01 - August 31: Each team completes the task within the specified time and submits the results and models (the number of formal submissions does not exceed two, the deadline is 23:59 on August 31; the results should be saved in the prescribed format, and the submission address will be Noticed in the form of email; the model should be accompanied by a detailed description document, including the framework used, dependent environment, etc.; it is recommended that users submit a version in advance (not as a formal submission) before submitting the first version of the formal model to ensure the normal running of the test environment) .
  • September 01 - September 24: The organizing committee evaluates the model submitted by the user based on the test data set, and ranks the teams.
  • September 25: Announce the results of the competition.
  • October 16: Hold the seminar.

6. Other information

Visit the competition website for details: https://lsprc.github.io/